Mobile Robot Target Recognition Problem Based On Multi-source Information Fusion

The basis for mobile robot to perform various tasks is to use various sensors to obtain real-time and accurate information about its environment.Although a certain type of sensor can realize the perception of environmental information to a certain extent,its mode is single and one-sided,and it is easily affected by cost,its own defects,external conditions and other factors,so it cannot fully and effectively express the environmental information required by the robot,and its application scenarios are extremely limited.To solve this problem,mobile robots are equipped with an increasing number of various sensors(hardware upgrades),or use a variety of fusion technologies(software upgrades),to meet the challenges of uncertain task requirements.Around the above problems,this paper selects two representative sensors,millimeter wave radar and camera,respectively to study the target recognition methods based on vision and millimeter wave radar.Accordingly,the target recognition methods based on multi-source information fusion are further explored,and experimental verification and analysis are carried out.The main research content of this paper is as follows.(1)In view of the low accuracy and poor real-time performance of visual sensors in the process of car and pedestrian recognition,this essay selects cameras as specific verification objects and proposes a deep convolutional neural network algorithm based on the improved YOLOv3 model.Firstly,the prediction layer of the model structure was increased from three to four layers.The image pyramid structure was used to obtain the feature maps of cars and pedestrians at different scales.The shallow features of multiple scales were fused to extract the contour details of cars and pedestrians.Secondly,Focal Loss function was introduced into the loss function of the model to solve the imbalance problem of positive and negative samples,reduce the weight of easily classified samples,and reduce the number of missed targets.Then,the K-medians clustering algorithm was used to replace the K-means clustering algorithm in the original YOLOv3 model to improve the robustness of the detection model.Finally,experiments are conducted to verify that the proposed method can improve the accuracy of the mobile robot’s vision system in recognizing cars and pedestrians,and meet the real-time requirements.(2)As the inherent data characteristics of millimeter wave radar lead to the unsatisfactory target recognition effect,and the problem of missing detection is easy to occur in pedestrian target recognition,this paper selects the millimeter wave radar sensor as the specific verification object,and proposes the processing method of millimeter wave radar target information based on convolutional neural network.Firstly,the characteristics of millimeter wave radar signals generated by different objects are divided into three categories.According to these categories of signals,invalid targets are eliminated by preprocessing method,and the target information of preliminary processing is obtained.Secondly,based on the deep learning theory,the convolutional neural network is designed for secondary processing of the previously obtained target information to obtain the image target frame of the millimeter-wave radar data information.Finally,experiments are carried out to verify that the proposed method can identify vehicle and pedestrian targets through millimeter wave radar sensors,and extract the corresponding image frame and distance information.(3)As the existing fusion strategy fails to make full use of millimeter wave radar information,the traditional image processing method is used in the visual recognition end,which leads to the low accuracy of target recognition.In this paper,millimeter wave radar and camera are selected as the concrete verification object,and a multi-source information fusion target recognition model is proposed.Firstly,the hierarchical architecture of multi-sensor fusion is selected according to the data processing mode of multi-sensor fusion strategy.Secondly,the information of the two sensors is synchronized in space and time to complete the coordinate conversion between them.Then,based on the deep learning theory,a multi-sensor fusion target recognition neural network model is designed to integrate the image frame information provided by the millimeter-wave radar and the image information provided by the camera.Finally,the millimeter-wave radar and image data set are established,and the model is trained and tested.Experiments are conducted to verify the effectiveness of the proposed method in improving the accuracy and robustness of target recognition.